pith. sign in

arxiv: 2605.04744 · v1 · submitted 2026-05-06 · 💻 cs.LG

MixINN: Accelerating Plant Breeding by Combining Mixed Models and Deep Learning for Interaction Prediction

Aike Potze , Fred van Eeuwijk , Ioannis N. Athanasiadis This is my paper

Pith reviewed 2026-05-08 18:11 UTC · model grok-4.3

classification 💻 cs.LG
keywords plant breedinggenotype-environment interactionmixed modelsdeep learningyield predictionmulti-environment trialscornclimate adaptation
0
0 comments X

The pith

MixINN first extracts genotype-environment interaction labels from historical trial data using mixed models and then trains a deep neural network to predict those interactions for new crop varieties under future conditions.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper aims to show that this two-step process yields better rankings of corn genotypes by expected yield than standard breeding methods. A sympathetic reader would care because climate change is shifting which varieties perform best in which places, and faster, more accurate selection could help maintain or increase food production. The evaluation on a large continental U.S. corn dataset indicates that selecting the top 20 percent of genotypes according to MixINN predictions delivers 5.8 percent higher average yield, rising to 7.2 percent when the target environment is specified. These gains are presented as competitive with real breeding programs. If the approach generalizes, it would shorten the cycle of testing and crossing needed to develop varieties suited to altered climates.

Core claim

The central claim is that isolating high-quality genotype-environment interaction labels with mixed models from historical multi-environment trials supplies training targets that allow a deep neural network to predict interactions for new genotypes in future environmental conditions, thereby improving genotype ranking and selection outcomes over current plant breeding practice.

What carries the argument

MixINN, a hybrid pipeline that first applies mixed models to historical data to isolate genotype-environment interaction labels and then uses those labels as targets for a deep neural network to predict interactions for unseen genotypes and altered environments.

If this is right

  • Improved ranking accuracy directly raises the average yield of the genotypes advanced to the next breeding cycle.
  • Performance gains increase when predictions are conditioned on specific target environments rather than broad averages.
  • The method shortens the time required to identify climate-adapted varieties by reducing the number of field trials needed for reliable ranking.
  • The same pipeline can be applied to other crops that have accumulated multi-environment trial records.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The approach could be combined with genomic markers to further reduce the need for extensive phenotyping.
  • Repeated application across breeding cycles might accumulate larger gains than single-cycle improvements suggest.
  • If climate shifts are gradual, historical interaction labels may become less representative, requiring periodic retraining on newer data.
  • Extension to non-corn species would test whether the mixed-model-plus-network separation of signal from noise is crop-agnostic.

Load-bearing premise

The genotype-environment interaction labels obtained from mixed models on past data contain enough reliable signal for a neural network to generalize accurately to new genotypes and to environmental conditions that have not yet been observed.

What would settle it

A new multi-environment trial with fresh genotypes and at least one previously unseen location or year in which the top-ranked genotypes selected by MixINN predictions do not produce higher measured yields than those selected by existing mixed-model or genomic-prediction baselines.

Figures

Figures reproduced from arXiv: 2605.04744 by Aike Potze, Fred van Eeuwijk, Ioannis N. Athanasiadis.

Figure 1
Figure 1. Figure 1: Increase in average plot yields when only corn genotypes with highest predicted yields are grown, plotted versus percentage selected view at source ↗
read the original abstract

Plant breeding underpins global food security through incremental, accumulating improvements in crop yield, quality and sustainability, achieved via repeated cycles of crop ranking, selection and crossing. Climate change disrupts this process by altering local growing conditions, thereby shifting the relative performance of crop genotypes. Predicting these relative changes in yield is critical for food security. Yet, this problem remains an open challenge in plant breeding, and relatively unexplored within the AI community. We propose MixINN, an approach that first isolates high-quality genotype-environment interaction labels using mixed models, and then predicts these interactions for new crop varieties in future environmental conditions with a deep neural network. We evaluate our method on a corn multi-environment trial across the continental United States and show improved prediction of genotype ranking over current plant breeding methods. MixINN demonstrated superior performance in identifying the 20% most productive corn genotypes, leading to a 5.8% higher average yield, which further improved to 7.2% when targeting specific growing environments. These are competitive results for real-world breeding programs, demonstrating the potential of AI research in accelerating the development of climate-adapted crops, and improving future food security under climate change.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes MixINN, a hybrid approach that first uses mixed models to isolate high-quality genotype-environment (GxE) interaction labels from multi-environment trial data, then trains a deep neural network to predict these interactions for new crop varieties under new environmental conditions. Evaluated on a real multi-environment corn trial dataset across the continental US, the method reports superior performance in identifying the top 20% most productive genotypes, with 5.8% higher average yield overall and 7.2% when targeting specific growing environments, positioning it as a tool to accelerate breeding for climate-adapted crops.

Significance. If the generalization claims hold, the work offers a practical integration of established statistical genetics tools with deep learning for GxE prediction, with concrete performance numbers on real breeding data that could inform selection decisions. The focus on ranking improvements rather than raw accuracy and the use of mixed-model-derived labels as supervision are sensible design choices that align with breeding practice. However, the absence of explicit tests for extrapolation beyond the observed data distribution substantially tempers the significance for the climate-change use case highlighted in the abstract.

major comments (2)
  1. [Abstract / Evaluation] Abstract and evaluation description: The central claim that MixINN enables prediction 'for new crop varieties in future environmental conditions' and improves breeding 'under climate change' is not supported by the reported experiments, which use only historical US corn trial data without temporal hold-out sets, spatial extrapolation, or augmentation with shifted climate covariates (e.g., altered temperature or precipitation). This leaves open whether the reported ranking gains reflect interpolation within observed GxE patterns rather than robustness to out-of-distribution conditions.
  2. [Evaluation] Evaluation: The headline performance figures (5.8% and 7.2% yield gains on top-20% genotypes) are presented without any description of the baseline methods used for comparison, statistical significance tests, cross-validation scheme, or safeguards against data leakage between the mixed-model label generation and the neural-network training phases. These omissions make it impossible to verify that the gains are attributable to the proposed hybrid architecture rather than implementation details or dataset artifacts.
minor comments (1)
  1. [Abstract] The abstract states that results are 'competitive for real-world breeding programs' but provides no quantitative reference to typical selection gains or industry benchmarks; adding such context would strengthen the interpretation of the 5.8–7.2% figures.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive review. We respond to each major comment below and have made revisions to the manuscript as indicated.

read point-by-point responses

  1. Referee: [Abstract / Evaluation] Abstract and evaluation description: The central claim that MixINN enables prediction 'for new crop varieties in future environmental conditions' and improves breeding 'under climate change' is not supported by the reported experiments, which use only historical US corn trial data without temporal hold-out sets, spatial extrapolation, or augmentation with shifted climate covariates (e.g., altered temperature or precipitation). This leaves open whether the reported ranking gains reflect interpolation within observed GxE patterns rather than robustness to out-of-distribution conditions.

    Authors: We acknowledge the validity of this observation. Our experiments are performed on historical data from US corn trials spanning multiple years and locations, which include environmental variation but no explicit future climate simulations or temporal splits. The reported gains demonstrate improved ranking within the observed distribution. To align the claims with the evidence, we have revised the abstract and introduction to focus on prediction for new varieties under new but observed environmental conditions, removing direct references to climate change adaptation. We have added a dedicated limitations section that discusses the challenges of OOD generalization and outlines future directions involving climate projections. This revision clarifies the scope without altering the empirical results. revision: yes

  2. Referee: [Evaluation] Evaluation: The headline performance figures (5.8% and 7.2% yield gains on top-20% genotypes) are presented without any description of the baseline methods used for comparison, statistical significance tests, cross-validation scheme, or safeguards against data leakage between the mixed-model label generation and the neural-network training phases. These omissions make it impossible to verify that the gains are attributable to the proposed hybrid architecture rather than implementation details or dataset artifacts.

    Authors: We apologize for insufficient detail in the initial submission. The full paper specifies baselines including standard mixed models (e.g., GBLUP for GxE) and competing DL methods, uses a cross-validation scheme based on holding out entire environments to mimic prediction to new locations, and generates mixed-model labels within each training fold to avoid leakage. We have now substantially expanded the Methods section with precise descriptions of all baselines, the CV procedure (including number of folds and environment selection), statistical analysis (paired tests on yield improvements across multiple runs), and a step-by-step pipeline ensuring separation of label generation and model training. These additions allow independent verification of the performance gains. revision: yes

Circularity Check

0 steps flagged

No circularity: MixINN uses standard mixed-model labeling followed by independent DNN prediction without self-referential reduction.

full rationale

The described pipeline isolates GxE labels via mixed models on historical trials then trains a separate neural network to predict interactions for new genotypes. This is a conventional two-stage ML workflow with no equation or claim reducing by construction to its own fitted inputs, no load-bearing self-citation for uniqueness, and no ansatz or renaming that collapses the result to the training labels. Evaluation metrics are computed on held-out portions of the same historical dataset, which is independent of the method definition itself. The paper therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract provides no explicit free parameters, axioms, or invented entities; the approach relies on standard mixed-model statistics and neural-network training without introducing new postulated objects.

pith-pipeline@v0.9.0 · 5514 in / 1179 out tokens · 51443 ms · 2026-05-08T18:11:44.312827+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

55 extracted references · 3 canonical work pages

  1. [1]

    Genomic selection in plant breeding: Key factors shaping two decades of progress.Molecular Plant, 17(4):552–578,

    [Alemuet al., 2024 ] Admas Alemu, Johanna ˚Astrand, Os- val A Montesinos-Lopez, Julio Isidro y Sanchez, Javier Fernandez-Gonzalez, Wuletaw Tadesse, Ramesh R Ve- tukuri, Anders S Carlsson, Alf Ceplitis, Jose Crossa, et al. Genomic selection in plant breeding: Key factors shaping two decades of progress.Molecular Plant, 17(4):552–578,

    2024

  2. [2]

    Layer Normalization

    [Baet al., 2016 ] Jimmy Lei Ba, Jamie Ryan Kiros, and Ge- offrey E Hinton. Layer normalization.arXiv preprint arXiv:1607.06450,

    work page Pith review arXiv 2016

  3. [3]

    Prediction of maize single- cross performance using rflps and information from related hybrids.Crop Science, 34(1):20–25,

    [Bernardo, 1994] Rex Bernardo. Prediction of maize single- cross performance using rflps and information from related hybrids.Crop Science, 34(1):20–25,

    1994

  4. [4]

    [Boeret al., 2007 ] Martin P Boer, Deanne Wright, Lizhi Feng, Dean W Podlich, Lang Luo, Mark Cooper, and Fred A Van Eeuwijk. A mixed-model quantitative trait loci (qtl) analysis for multiple-environment trial data using environmental covariables for qtl-by-environment interac- tions, with an example in maize.Genetics, 177(3):1801– 1813,

    2007

  5. [5]

    Random forests.Machine learning, 45(1):5–32,

    [Breiman, 2001] Leo Breiman. Random forests.Machine learning, 45(1):5–32,

    2001

  6. [6]

    Asreml-r reference man- ual version 4.VSN International Ltd, Hemel Hempstead, HP1 1ES, UK,

    [Butleret al., 2017 ] DG Butler, BR Cullis, AR Gilmour, BJ Gogel, and Ret al Thompson. Asreml-r reference man- ual version 4.VSN International Ltd, Hemel Hempstead, HP1 1ES, UK,

    2017

  7. [7]

    Rubi: Reducing uni- modal biases for visual question answering.Advances in neural information processing systems, 32,

    [Cadeneet al., 2019 ] Remi Cadene, Corentin Dancette, Matthieu Cord, Devi Parikh, et al. Rubi: Reducing uni- modal biases for visual question answering.Advances in neural information processing systems, 32,

    2019

  8. [8]

    comet.ml

    [Comet ML Inc., 2023] Comet ML Inc. comet.ml. https:// pypi.org/project/comet-ml/3.34.0/,

    2023

  9. [9]

    [Condonet al., 2004 ] Anthony G Condon, RA Richards, GJ Rebetzke, and GD Farquhar

    Version 3.34.0. [Condonet al., 2004 ] Anthony G Condon, RA Richards, GJ Rebetzke, and GD Farquhar. Breeding for high water-use efficiency.Journal of experimental botany, 55(407):2447–2460,

    2004

  10. [10]

    Use of crop growth models with whole-genome prediction: application to a maize multienvironment trial.Crop Science, 56(5):2141– 2156,

    [Cooperet al., 2016 ] Mark Cooper, Frank Technow, Carlos Messina, Carla Gho, and L Radu Totir. Use of crop growth models with whole-genome prediction: application to a maize multienvironment trial.Crop Science, 56(5):2141– 2156,

    2016

  11. [11]

    Extending the breeder’s equation to take aim at the target population of environments.Frontiers in Plant Science, 14:1129591,

    [Cooperet al., 2023 ] Mark Cooper, Owen Powell, Carla Gho, Tom Tang, and Carlos Messina. Extending the breeder’s equation to take aim at the target population of environments.Frontiers in Plant Science, 14:1129591,

    2023

  12. [12]

    Deep neural networks for youtube recom- mendations

    [Covingtonet al., 2016 ] Paul Covington, Jay Adams, and Emre Sargin. Deep neural networks for youtube recom- mendations. InProceedings of the 10th ACM conference on recommender systems, pages 191–198,

    2016

  13. [13]

    Machine learning algorithms translate big data into predictive breeding accuracy.Trends in Plant Science, 30(2):167–184,

    [Crossaet al., 2025 ] Jos´e Crossa, Osval A Montesinos- Lopez, Germano Costa-Neto, Paolo Vitale, Johannes WR Martini, Daniel Runcie, Roberto Fritsche-Neto, Abelardo Montesinos-Lopez, Paulino P ´erez-Rodr´ıguez, Guillermo Gerard, et al. Machine learning algorithms translate big data into predictive breeding accuracy.Trends in Plant Science, 30(2):167–184,

    2025

  14. [14]

    Half a century of studying genotype×environment interactions in plant breeding ex- periments.Crop Science, 56(5):2090–2105,

    [Eliaset al., 2016 ] Ani A Elias, Kelly R Robbins, RW Do- erge, and Mitchell R Tuinstra. Half a century of studying genotype×environment interactions in plant breeding ex- periments.Crop Science, 56(5):2090–2105,

    2016

  15. [15]

    The analysis of adaptation in a plant-breeding pro- gramme.Australian journal of agricultural research, 14(6):742–754,

    [Finlay and Wilkinson, 1963] KW Finlay and GN Wilkin- son. The analysis of adaptation in a plant-breeding pro- gramme.Australian journal of agricultural research, 14(6):742–754,

    1963

  16. [16]

    Crop yields and global food security.ACIAR: Canberra, ACT, pages 8–11,

    [Fischeret al., 2014 ] RA Fischer, Derek Byerlee, and Greg Edmeades. Crop yields and global food security.ACIAR: Canberra, ACT, pages 8–11,

    2014

  17. [17]

    [Gauch, 1992] Jr Gauch, HG.Statistical analysis of regional yield trials: AMMI analysis of factorial designs.Elsevier,

    1992

  18. [18]

    Megalmm improves genomic predictions in new environments using environmental covariates.Genetics, 229(1):1–41,

    [Huet al., 2025 ] Haixiao Hu, Renaud Rincent, and Daniel E Runcie. Megalmm improves genomic predictions in new environments using environmental covariates.Genetics, 229(1):1–41,

    2025

  19. [19]

    A reaction norm model for genomic se- lection using high-dimensional genomic and environmen- tal data.Theoretical and applied genetics, 127:595–607,

    [Jarqu´ınet al., 2014] Diego Jarqu´ın, Jos´e Crossa, Xavier La- caze, Philippe Du Cheyron, Jo ¨elle Daucourt, Josiane Lor- geou, Franc ¸ois Piraux, Laurent Guerreiro, Paulino P´erez, Mario Calus, et al. A reaction norm model for genomic se- lection using high-dimensional genomic and environmen- tal data.Theoretical and applied genetics, 127:595–607,

    2014

  20. [20]

    Crop yield prediction using deep neural networks.Frontiers in plant science, 10:621,

    [Khaki and Wang, 2019] Saeed Khaki and Lizhi Wang. Crop yield prediction using deep neural networks.Frontiers in plant science, 10:621,

    2019

  21. [21]

    Yield prediction through inte- gration of genetic, environment, and management data through deep learning.G3: Genes, Genomes, Genetics, 13(4):jkad006,

    [Kicket al., 2023 ] Daniel R Kick, Jason G Wallace, James C Schnable, Judith M Kolkman, Barıs ¸ Alaca, Timothy M Beissinger, Jode Edwards, David Ertl, Sherry Flint-Garcia, Joseph L Gage, et al. Yield prediction through inte- gration of genetic, environment, and management data through deep learning.G3: Genes, Genomes, Genetics, 13(4):jkad006,

    2023

  22. [22]

    Diverse concepts of breeding for nitrogen use efficiency

    [Lammerts van Bueren and Struik, 2017] Edith T Lam- merts van Bueren and Paul C Struik. Diverse concepts of breeding for nitrogen use efficiency. a review.Agronomy for Sustainable Development, 37(5):50,

    2017

  23. [23]

    Backpropagation ap- plied to handwritten zip code recognition.Neural compu- tation, 1(4):541–551,

    [LeCunet al., 1989 ] Yann LeCun, Bernhard Boser, John S Denker, Donnie Henderson, Richard E Howard, Wayne Hubbard, and Lawrence D Jackel. Backpropagation ap- plied to handwritten zip code recognition.Neural compu- tation, 1(4):541–551,

    1989

  24. [24]

    Genomes to fields 2022 maize genotype by environment prediction competi- tion.BMC Research Notes, 16(1):148,

    [Limaet al., 2023 ] Dayane Cristina Lima, Jacob D Wash- burn, Jos ´e Ignacio Varela, Qiuyue Chen, Joseph L Gage, Maria Cinta Romay, James Holland, David Ertl, Marco Lopez-Cruz, Fernando M Aguate, et al. Genomes to fields 2022 maize genotype by environment prediction competi- tion.BMC Research Notes, 16(1):148,

    2023

  25. [25]

    Decoupled weight decay regularization,

    [Loshchilov and Hutter, 2019] Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization,

    2019

  26. [26]

    The statistical analy- sis of multi-environment data: modeling genotype-by- environment interaction and its genetic basis.Frontiers in physiology, 4:44,

    [Malosettiet al., 2013 ] Marcos Malosetti, Jean-Marcel Rib- aut, and Fred A van Eeuwijk. The statistical analy- sis of multi-environment data: modeling genotype-by- environment interaction and its genetic basis.Frontiers in physiology, 4:44,

    2013

  27. [27]

    Prediction of total genetic value using genome-wide dense marker maps.genetics, 157(4):1819–1829,

    [Meuwissenet al., 2001 ] Theo HE Meuwissen, Ben J Hayes, and ME1461589 Goddard. Prediction of total genetic value using genome-wide dense marker maps.genetics, 157(4):1819–1829,

    2001

  28. [28]

    Rectified linear units improve restricted boltzmann ma- chines

    [Nair and Hinton, 2010] Vinod Nair and Geoffrey E Hinton. Rectified linear units improve restricted boltzmann ma- chines. InProceedings of the 27th international con- ference on machine learning (ICML-10), pages 807–814,

    2010

  29. [29]

    Gene-by-environment interactions in plants: Molecular mechanisms, environmental drivers, and adaptive plasticity.The Plant Cell, 35(1):109–124,

    [Napieret al., 2023 ] Joseph D Napier, Robert W Heckman, and Thomas E Juenger. Gene-by-environment interactions in plants: Molecular mechanisms, environmental drivers, and adaptive plasticity.The Plant Cell, 35(1):109–124,

    2023

  30. [30]

    Climax: A foundation model for weather and climate.arXiv preprint arXiv:2301.10343, 2023

    [Nguyenet al., 2023 ] Tung Nguyen, Johannes Brandstetter, Ashish Kapoor, Jayesh K Gupta, and Aditya Grover. Cli- max: A foundation model for weather and climate.arXiv preprint arXiv:2301.10343,

    work page arXiv 2023

  31. [31]

    Recovery of inter-block information when block sizes are unequal.Biometrika, 58(3):545–554,

    [Patterson and Thompson, 1971] H Desmond Patterson and Robin Thompson. Recovery of inter-block information when block sizes are unequal.Biometrika, 58(3):545–554,

    1971

  32. [32]

    Factor-analytic variance–covariance structures for prediction into a target population of environments

    [Piepho and Williams, 2024] Hans-Peter Piepho and Emlyn Williams. Factor-analytic variance–covariance structures for prediction into a target population of environments. Biometrical Journal, 66(6):e202400008,

    2024

  33. [33]

    Blup for phenotypic selection in plant breeding and variety testing.Euphytica, 161(1):209–228,

    [Piephoet al., 2008 ] HP Piepho, J M¨ohring, AE Melchinger, and A B ¨uchse. Blup for phenotypic selection in plant breeding and variety testing.Euphytica, 161(1):209–228,

    2008

  34. [34]

    Analyzing genotype- environment data by mixed models with multiplicative terms.Biometrics, pages 761–766,

    [Piepho, 1997] H-P Piepho. Analyzing genotype- environment data by mixed models with multiplicative terms.Biometrics, pages 761–766,

    1997

  35. [35]

    Empirical best linear un- biased prediction in cultivar trials using factor-analytic variance-covariance structures.Theoretical and Applied Genetics, 97(1):195–201,

    [Piepho, 1998] Hans-Peter Piepho. Empirical best linear un- biased prediction in cultivar trials using factor-analytic variance-covariance structures.Theoretical and Applied Genetics, 97(1):195–201,

    1998

  36. [36]

    Structured multimodal deep learn- ing improves genomic prediction in future environments

    [Potzeet al., 2025 ] Aike Potze, Fred van Eeuwijk, and Ioan- nis N Athanasiadis. Structured multimodal deep learn- ing improves genomic prediction in future environments. bioRxiv, pages 2025–09,

    2025

  37. [37]

    Evalua- tion and optimization of wheat and maize national eval- uation systems in europe.Preprint (Version

    [Ramakerset al., 2025 ] Jip Ramakers, Waqas Malik, Claude Welcker, Boris Parent, Daniela Bustos-Korts, Nadir Abu-Samra Spencer, Harimurti Buntaran, Xinxin Chen, Damien Eylenbosch, Clemens Flamm, et al. Evalua- tion and optimization of wheat and maize national eval- uation systems in europe.Preprint (Version

    2025

  38. [38]

    avail- able at Research Square [https://doi.org/10.21203/rs.3.rs- 6221922/v1],

    work page doi:10.21203/rs.3.rs-

  39. [39]

    Gaussian pro- cesses in machine learning

    [Rasmussen, 2003] Carl Edward Rasmussen. Gaussian pro- cesses in machine learning. InSummer school on machine learning, pages 63–71. Springer,

    2003

  40. [40]

    That blup is a good thing: the estimation of random effects.Statistical science, pages 15–32,

    [Robinson, 1991] George K Robinson. That blup is a good thing: the estimation of random effects.Statistical science, pages 15–32,

    1991

  41. [41]

    [Robinson, 1996] Raoul A Robinson.Return to resistance: breeding crops to reduce pesticide dependence. IDRC,

    1996

  42. [42]

    qgg: an r package for large-scale quantitative genetic analyses

    [Rohdeet al., 2020 ] Palle Duun Rohde, Izel Fourie Sørensen, and Peter Sørensen. qgg: an r package for large-scale quantitative genetic analyses. Bioinformatics, 36(8):2614–2615,

    2020

  43. [43]

    Genotype by environment interaction and adaptation

    [Romagosaet al., 2013 ] Ignacio Romagosa, Gisela Borr `as- Gelonch, Gustavo Slafer, and Fred van Eeuwijk. Genotype by environment interaction and adaptation. InSustainable Food Production, pages 846–870. Springer,

    2013

  44. [44]

    Learning representations by back-propagating errors.nature, 323(6088):533–536,

    [Rumelhartet al., 1986 ] David E Rumelhart, Geoffrey E Hinton, and Ronald J Williams. Learning representations by back-propagating errors.nature, 323(6088):533–536,

    1986

  45. [45]

    Integrating random effects in deep neural networks.Journal of Machine Learning Research, 24(156):1–57,

    [Simchoni and Rosset, 2023] Giora Simchoni and Saharon Rosset. Integrating random effects in deep neural networks.Journal of Machine Learning Research, 24(156):1–57,

    2023

  46. [46]

    Analyzing variety by environment data using multiplicative mixed models and adjustments for spatial field trend.Biometrics, 57(4):1138–1147,

    [Smithet al., 2001 ] Alison Smith, Brian Cullis, and Robin Thompson. Analyzing variety by environment data using multiplicative mixed models and adjustments for spatial field trend.Biometrics, 57(4):1138–1147,

    2001

  47. [47]

    Dropout: a simple way to prevent neural networks from overfitting.The journal of machine learning re- search, 15(1):1929–1958,

    [Srivastavaet al., 2014 ] Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdi- nov. Dropout: a simple way to prevent neural networks from overfitting.The journal of machine learning re- search, 15(1):1929–1958,

    2014

  48. [48]

    The lamb- daloss framework for ranking metric optimization

    [Wanget al., 2018 ] Xuanhui Wang, Cheng Li, Nadav Gol- bandi, Michael Bendersky, and Marc Najork. The lamb- daloss framework for ranking metric optimization. InPro- ceedings of the 27th ACM international conference on in- formation and knowledge management, pages 1313–1322,

    2018

  49. [49]

    Predicting phenotypes from genetic, en- vironment, management, and historical data using cnns

    [Washburnet al., 2021 ] Jacob D Washburn, Emre Cimen, Guillaume Ramstein, Timothy Reeves, Patrick O’Briant, Greg McLean, Mark Cooper, Graeme Hammer, and Ed- ward S Buckler. Predicting phenotypes from genetic, en- vironment, management, and historical data using cnns. Theoretical and Applied Genetics, 134(12):3997–4011,

    2021

  50. [50]

    Global genotype by environment prediction competition reveals that diverse modeling strategies can deliver satisfactory maize yield estimates.Genetics, 229(2),

    [Washburnet al., 2025 ] Jacob D Washburn, Jos ´e Ignacio Varela, Alencar Xavier, Qiuyue Chen, David Ertl, Joseph L Gage, James B Holland, Dayane Cristina Lima, Maria Cinta Romay, Marco Lopez-Cruz, et al. Global genotype by environment prediction competition reveals that diverse modeling strategies can deliver satisfactory maize yield estimates.Genetics, 229(2),

    2025

  51. [51]

    Enhancing genetic gain in the era of molecular breeding.Journal of Experimental Botany, 68(11):2641–2666,

    [Xuet al., 2017 ] Yunbi Xu, Ping Li, Cheng Zou, Yanli Lu, Chuanxiao Xie, Xuecai Zhang, Boddupalli M Prasanna, and Michael S Olsen. Enhancing genetic gain in the era of molecular breeding.Journal of Experimental Botany, 68(11):2641–2666,

    2017

  52. [52]

    CRC press,

    [Yan and Kang, 2002] Weikai Yan and Manjit S Kang.GGE biplot analysis: A graphical tool for breeders, geneticists, and agronomists. CRC press,

    2002

  53. [53]

    Deep gaussian process for crop yield prediction based on remote sensing data

    [Youet al., 2017 ] Jiaxuan You, Xiaocheng Li, Melvin Low, David Lobell, and Stefano Ermon. Deep gaussian process for crop yield prediction based on remote sensing data. In Proceedings of the AAAI conference on artificial intelli- gence, volume 31,

    2017

  54. [54]

    7 Appendix A. Dataset filtering and imputation In this section, we describe our preprocessing steps for the Genomes to Fields 2022 Maize G×E Prediction Challenge [Limaet al., 2023 ] dataset, filtering of samples and impu- tation of features. The dataset was divided into two sets: a training set (years 2015-2021) and a test set (year 2022). Filtering of sa...

    2022

  55. [55]

    No early stopping was used

    From the given ranges, we sampled random configurations to tune. No early stopping was used. G2F-DNN [Kicket al., 2023 ] was trained on crop yields, and only validation samples with both unseen genotypes and unseen environments were used to evaluate each model. The architectural hyperparameters were kept fixed, as the origi- nal work reported extensive tu...

    2023